github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-07 03:00:26 -04:00
Code Issues Packages Projects Releases Wiki Activity

CLANG doesn't support float4

Browse Source
This commit is contained in:
George Hotz
2023-02-22 10:41:15 -08:00
parent c4c2c28738
commit a3ddc1d484
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tinygrad/llops/ops_gpu.py
View File

@@ -59,7 +59,7 @@ class CLASTKernel(ASTKernel):
assert len(value) == self.buftokens[buf_index].size(), f"size mismatch {len(value)} != {self.buftokens[buf_index].size()}"
can_merge = (not self.bufs[buf_index].st.needs_valid() and len(self.bufs[buf_index].st.views) == 1) or "Image" in str(type(self.bufs[buf_index]._buf))
should_upcast = can_merge and self.buftokens[buf_index].can_float4()
should_upcast = not CLANG and can_merge and self.buftokens[buf_index].can_float4()
to_store = {o:v for o,v in zip(self.buftokens[buf_index].offsets(), value)}
did_store = set()
@@ -88,7 +88,7 @@ class CLASTKernel(ASTKernel):
const = Token(f"({self.bufs[buf_index]._backing[0]}f)", Types.FLOAT)
can_merge = (not self.bufs[buf_index].st.needs_valid() and len(self.bufs[buf_index].st.views) == 1) or "Image" in str(type(self.bufs[buf_index]._buf))
should_upcast = const is None and can_merge and self.buftokens[buf_index].can_float4()
should_upcast = not CLANG and const is None and can_merge and self.buftokens[buf_index].can_float4()
tokens = []
for o in self.buftokens[buf_index].offsets():